Externally Trained Classifiers

AbstractSequence-to-sequence models for open-domain dialogue generation tend to favor generic, uninformative responses. Past work has focused on word frequency-based approaches to improving specificity, such as penalizing responses with only common …...